Predictive model and risk engine web application for surgical site infection risk in perioperative patients with type 2 diabetes

Aim To identify predictive factors for surgical site infection (SSI) in patients with type 2 diabetes and develop a prediction tool. Materials and methods We retrospectively analyzed the perioperative blood glucose management of 105 patients with type 2 diabetes treated from 2016 to 2018 at Chiba University Hospital. The primary outcome was SSI onset within 30 postoperative days; moreover, predictive factors were identified using univariate analysis. Principal component analysis and logistic regression analysis were performed to prepare SSI predictive model using the identified predictive factors. The area under the receiver operating characteristic curve (AUC) was evaluated. Based on the predictive model, we developed a risk engine for SSI prediction. Results Compared with patients without SSI (n = 70), those with SSI (n= 35) had significantly higher fasting blood glucose levels at referral (169.1 +/- 61.8 mg/dL vs. 140.1 +/- 56.6, P = 0.036), preoperative mean blood glucose levels (178.3 +/- 48.4 mg/ dL vs. 155.2 +/- 39.7, P = 0.009), preoperative maximum blood glucose levels (280.4 +/- 87.3 mg/dL vs. 230.3 +/- 92.4, P = 0.009), preoperative blood glucose fluctuations (54.9 +/- 24.1 mg/dL vs. 37.7 +/- 23.1, P = 0.001), percentage of hospitalization at referral (54.3% vs. 20.0, P < 0.001); longer operation time (432.5 +/- 179.6 min vs. 282.5 +/- 178.3, P < 0.001); and greater bleeding volume (972.3 +/- 920.1 mg/dL vs. 436.4 +/- 795.8, P < 0.001). Logistic regression analysis revealed preoperative blood glucose fluctuation and operation time as the most reliable predictive factors. The predictive model had high prediction accuracy (AUC of 0.801). The risk engine prototype for SSI prediction can be accessed at https://www.dm-ope-riskengine.org/. Conclusions The predictive model developed in this study could screen high-risk patients. It may be useful to prevent SSI in such patients.